2 * linux/fs/ext2/ialloc.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
9 * BSD ufs-inspired inode and directory allocation by
10 * Stephen Tweedie (sct@dcs.ed.ac.uk), 1993
11 * Big-endian to little-endian byte-swapping/bitmaps by
12 * David S. Miller (davem@caip.rutgers.edu), 1995
15 #include <linux/quotaops.h>
16 #include <linux/sched.h>
17 #include <linux/backing-dev.h>
18 #include <linux/buffer_head.h>
19 #include <linux/random.h>
25 * ialloc.c contains the inodes allocation and deallocation routines
29 * The free inodes are managed by bitmaps. A file system contains several
30 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
31 * block for inodes, N blocks for the inode table and data blocks.
33 * The file system contains group descriptors which are located after the
34 * super block. Each descriptor contains the number of the bitmap block and
35 * the free blocks count in the block.
40 * Read the inode allocation bitmap for a given block_group, reading
41 * into the specified slot in the superblock's bitmap cache.
43 * Return buffer_head of bitmap on success or NULL.
45 static struct buffer_head
*
46 read_inode_bitmap(struct super_block
* sb
, unsigned long block_group
)
48 struct ext2_group_desc
*desc
;
49 struct buffer_head
*bh
= NULL
;
51 desc
= ext2_get_group_desc(sb
, block_group
, NULL
);
55 bh
= sb_bread(sb
, le32_to_cpu(desc
->bg_inode_bitmap
));
57 ext2_error(sb
, "read_inode_bitmap",
58 "Cannot read inode bitmap - "
59 "block_group = %lu, inode_bitmap = %u",
60 block_group
, le32_to_cpu(desc
->bg_inode_bitmap
));
65 static void ext2_release_inode(struct super_block
*sb
, int group
, int dir
)
67 struct ext2_group_desc
* desc
;
68 struct buffer_head
*bh
;
70 desc
= ext2_get_group_desc(sb
, group
, &bh
);
72 ext2_error(sb
, "ext2_release_inode",
73 "can't get descriptor for group %d", group
);
77 spin_lock(sb_bgl_lock(EXT2_SB(sb
), group
));
78 le16_add_cpu(&desc
->bg_free_inodes_count
, 1);
80 le16_add_cpu(&desc
->bg_used_dirs_count
, -1);
81 spin_unlock(sb_bgl_lock(EXT2_SB(sb
), group
));
83 percpu_counter_dec(&EXT2_SB(sb
)->s_dirs_counter
);
85 mark_buffer_dirty(bh
);
89 * NOTE! When we get the inode, we're the only people
90 * that have access to it, and as such there are no
91 * race conditions we have to worry about. The inode
92 * is not on the hash-lists, and it cannot be reached
93 * through the filesystem because the directory entry
94 * has been deleted earlier.
96 * HOWEVER: we must make sure that we get no aliases,
97 * which means that we have to call "clear_inode()"
98 * _before_ we mark the inode not in use in the inode
99 * bitmaps. Otherwise a newly created file might use
100 * the same inode number (not actually the same pointer
101 * though), and then we'd have two inodes sharing the
102 * same inode number and space on the harddisk.
104 void ext2_free_inode (struct inode
* inode
)
106 struct super_block
* sb
= inode
->i_sb
;
109 struct buffer_head
*bitmap_bh
;
110 unsigned long block_group
;
112 struct ext2_super_block
* es
;
115 ext2_debug ("freeing inode %lu\n", ino
);
118 * Note: we must free any quota before locking the superblock,
119 * as writing the quota to disk may need the lock as well.
121 if (!is_bad_inode(inode
)) {
122 /* Quota is already initialized in iput() */
123 ext2_xattr_delete_inode(inode
);
124 dquot_free_inode(inode
);
128 es
= EXT2_SB(sb
)->s_es
;
129 is_directory
= S_ISDIR(inode
->i_mode
);
131 /* Do this BEFORE marking the inode not in use or returning an error */
134 if (ino
< EXT2_FIRST_INO(sb
) ||
135 ino
> le32_to_cpu(es
->s_inodes_count
)) {
136 ext2_error (sb
, "ext2_free_inode",
137 "reserved or nonexistent inode %lu", ino
);
140 block_group
= (ino
- 1) / EXT2_INODES_PER_GROUP(sb
);
141 bit
= (ino
- 1) % EXT2_INODES_PER_GROUP(sb
);
142 bitmap_bh
= read_inode_bitmap(sb
, block_group
);
146 /* Ok, now we can actually update the inode bitmaps.. */
147 if (!ext2_clear_bit_atomic(sb_bgl_lock(EXT2_SB(sb
), block_group
),
148 bit
, (void *) bitmap_bh
->b_data
))
149 ext2_error (sb
, "ext2_free_inode",
150 "bit already cleared for inode %lu", ino
);
152 ext2_release_inode(sb
, block_group
, is_directory
);
153 mark_buffer_dirty(bitmap_bh
);
154 if (sb
->s_flags
& MS_SYNCHRONOUS
)
155 sync_dirty_buffer(bitmap_bh
);
161 * We perform asynchronous prereading of the new inode's inode block when
162 * we create the inode, in the expectation that the inode will be written
163 * back soon. There are two reasons:
165 * - When creating a large number of files, the async prereads will be
166 * nicely merged into large reads
167 * - When writing out a large number of inodes, we don't need to keep on
168 * stalling the writes while we read the inode block.
170 * FIXME: ext2_get_group_desc() needs to be simplified.
172 static void ext2_preread_inode(struct inode
*inode
)
174 unsigned long block_group
;
175 unsigned long offset
;
177 struct ext2_group_desc
* gdp
;
178 struct backing_dev_info
*bdi
;
180 bdi
= inode
->i_mapping
->backing_dev_info
;
181 if (bdi_read_congested(bdi
))
183 if (bdi_write_congested(bdi
))
186 block_group
= (inode
->i_ino
- 1) / EXT2_INODES_PER_GROUP(inode
->i_sb
);
187 gdp
= ext2_get_group_desc(inode
->i_sb
, block_group
, NULL
);
192 * Figure out the offset within the block group inode table
194 offset
= ((inode
->i_ino
- 1) % EXT2_INODES_PER_GROUP(inode
->i_sb
)) *
195 EXT2_INODE_SIZE(inode
->i_sb
);
196 block
= le32_to_cpu(gdp
->bg_inode_table
) +
197 (offset
>> EXT2_BLOCK_SIZE_BITS(inode
->i_sb
));
198 sb_breadahead(inode
->i_sb
, block
);
202 * There are two policies for allocating an inode. If the new inode is
203 * a directory, then a forward search is made for a block group with both
204 * free space and a low directory-to-inode ratio; if that fails, then of
205 * the groups with above-average free space, that group with the fewest
206 * directories already is chosen.
208 * For other inodes, search forward from the parent directory\'s block
209 * group to find a free inode.
211 static int find_group_dir(struct super_block
*sb
, struct inode
*parent
)
213 int ngroups
= EXT2_SB(sb
)->s_groups_count
;
214 int avefreei
= ext2_count_free_inodes(sb
) / ngroups
;
215 struct ext2_group_desc
*desc
, *best_desc
= NULL
;
216 int group
, best_group
= -1;
218 for (group
= 0; group
< ngroups
; group
++) {
219 desc
= ext2_get_group_desc (sb
, group
, NULL
);
220 if (!desc
|| !desc
->bg_free_inodes_count
)
222 if (le16_to_cpu(desc
->bg_free_inodes_count
) < avefreei
)
225 (le16_to_cpu(desc
->bg_free_blocks_count
) >
226 le16_to_cpu(best_desc
->bg_free_blocks_count
))) {
238 * Orlov's allocator for directories.
240 * We always try to spread first-level directories.
242 * If there are blockgroups with both free inodes and free blocks counts
243 * not worse than average we return one with smallest directory count.
244 * Otherwise we simply return a random group.
246 * For the rest rules look so:
248 * It's OK to put directory into a group unless
249 * it has too many directories already (max_dirs) or
250 * it has too few free inodes left (min_inodes) or
251 * it has too few free blocks left (min_blocks) or
252 * it's already running too large debt (max_debt).
253 * Parent's group is preferred, if it doesn't satisfy these
254 * conditions we search cyclically through the rest. If none
255 * of the groups look good we just look for a group with more
256 * free inodes than average (starting at parent's group).
258 * Debt is incremented each time we allocate a directory and decremented
259 * when we allocate an inode, within 0--255.
262 #define INODE_COST 64
263 #define BLOCK_COST 256
265 static int find_group_orlov(struct super_block
*sb
, struct inode
*parent
)
267 int parent_group
= EXT2_I(parent
)->i_block_group
;
268 struct ext2_sb_info
*sbi
= EXT2_SB(sb
);
269 struct ext2_super_block
*es
= sbi
->s_es
;
270 int ngroups
= sbi
->s_groups_count
;
271 int inodes_per_group
= EXT2_INODES_PER_GROUP(sb
);
278 int max_debt
, max_dirs
, min_blocks
, min_inodes
;
280 struct ext2_group_desc
*desc
;
282 freei
= percpu_counter_read_positive(&sbi
->s_freeinodes_counter
);
283 avefreei
= freei
/ ngroups
;
284 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
285 avefreeb
= free_blocks
/ ngroups
;
286 ndirs
= percpu_counter_read_positive(&sbi
->s_dirs_counter
);
288 if ((parent
== sb
->s_root
->d_inode
) ||
289 (EXT2_I(parent
)->i_flags
& EXT2_TOPDIR_FL
)) {
290 struct ext2_group_desc
*best_desc
= NULL
;
291 int best_ndir
= inodes_per_group
;
294 get_random_bytes(&group
, sizeof(group
));
295 parent_group
= (unsigned)group
% ngroups
;
296 for (i
= 0; i
< ngroups
; i
++) {
297 group
= (parent_group
+ i
) % ngroups
;
298 desc
= ext2_get_group_desc (sb
, group
, NULL
);
299 if (!desc
|| !desc
->bg_free_inodes_count
)
301 if (le16_to_cpu(desc
->bg_used_dirs_count
) >= best_ndir
)
303 if (le16_to_cpu(desc
->bg_free_inodes_count
) < avefreei
)
305 if (le16_to_cpu(desc
->bg_free_blocks_count
) < avefreeb
)
308 best_ndir
= le16_to_cpu(desc
->bg_used_dirs_count
);
311 if (best_group
>= 0) {
320 ndirs
= 1; /* percpu_counters are approximate... */
322 blocks_per_dir
= (le32_to_cpu(es
->s_blocks_count
)-free_blocks
) / ndirs
;
324 max_dirs
= ndirs
/ ngroups
+ inodes_per_group
/ 16;
325 min_inodes
= avefreei
- inodes_per_group
/ 4;
326 min_blocks
= avefreeb
- EXT2_BLOCKS_PER_GROUP(sb
) / 4;
328 max_debt
= EXT2_BLOCKS_PER_GROUP(sb
) / max(blocks_per_dir
, BLOCK_COST
);
329 if (max_debt
* INODE_COST
> inodes_per_group
)
330 max_debt
= inodes_per_group
/ INODE_COST
;
336 for (i
= 0; i
< ngroups
; i
++) {
337 group
= (parent_group
+ i
) % ngroups
;
338 desc
= ext2_get_group_desc (sb
, group
, NULL
);
339 if (!desc
|| !desc
->bg_free_inodes_count
)
341 if (sbi
->s_debts
[group
] >= max_debt
)
343 if (le16_to_cpu(desc
->bg_used_dirs_count
) >= max_dirs
)
345 if (le16_to_cpu(desc
->bg_free_inodes_count
) < min_inodes
)
347 if (le16_to_cpu(desc
->bg_free_blocks_count
) < min_blocks
)
353 for (i
= 0; i
< ngroups
; i
++) {
354 group
= (parent_group
+ i
) % ngroups
;
355 desc
= ext2_get_group_desc (sb
, group
, NULL
);
356 if (!desc
|| !desc
->bg_free_inodes_count
)
358 if (le16_to_cpu(desc
->bg_free_inodes_count
) >= avefreei
)
364 * The free-inodes counter is approximate, and for really small
365 * filesystems the above test can fail to find any blockgroups
377 static int find_group_other(struct super_block
*sb
, struct inode
*parent
)
379 int parent_group
= EXT2_I(parent
)->i_block_group
;
380 int ngroups
= EXT2_SB(sb
)->s_groups_count
;
381 struct ext2_group_desc
*desc
;
385 * Try to place the inode in its parent directory
387 group
= parent_group
;
388 desc
= ext2_get_group_desc (sb
, group
, NULL
);
389 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
) &&
390 le16_to_cpu(desc
->bg_free_blocks_count
))
394 * We're going to place this inode in a different blockgroup from its
395 * parent. We want to cause files in a common directory to all land in
396 * the same blockgroup. But we want files which are in a different
397 * directory which shares a blockgroup with our parent to land in a
398 * different blockgroup.
400 * So add our directory's i_ino into the starting point for the hash.
402 group
= (group
+ parent
->i_ino
) % ngroups
;
405 * Use a quadratic hash to find a group with a free inode and some
408 for (i
= 1; i
< ngroups
; i
<<= 1) {
410 if (group
>= ngroups
)
412 desc
= ext2_get_group_desc (sb
, group
, NULL
);
413 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
) &&
414 le16_to_cpu(desc
->bg_free_blocks_count
))
419 * That failed: try linear search for a free inode, even if that group
420 * has no free blocks.
422 group
= parent_group
;
423 for (i
= 0; i
< ngroups
; i
++) {
424 if (++group
>= ngroups
)
426 desc
= ext2_get_group_desc (sb
, group
, NULL
);
427 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
))
437 struct inode
*ext2_new_inode(struct inode
*dir
, int mode
)
439 struct super_block
*sb
;
440 struct buffer_head
*bitmap_bh
= NULL
;
441 struct buffer_head
*bh2
;
444 struct inode
* inode
;
445 struct ext2_group_desc
*gdp
;
446 struct ext2_super_block
*es
;
447 struct ext2_inode_info
*ei
;
448 struct ext2_sb_info
*sbi
;
452 inode
= new_inode(sb
);
454 return ERR_PTR(-ENOMEM
);
460 if (test_opt(sb
, OLDALLOC
))
461 group
= find_group_dir(sb
, dir
);
463 group
= find_group_orlov(sb
, dir
);
465 group
= find_group_other(sb
, dir
);
472 for (i
= 0; i
< sbi
->s_groups_count
; i
++) {
473 gdp
= ext2_get_group_desc(sb
, group
, &bh2
);
475 bitmap_bh
= read_inode_bitmap(sb
, group
);
482 repeat_in_this_group
:
483 ino
= ext2_find_next_zero_bit((unsigned long *)bitmap_bh
->b_data
,
484 EXT2_INODES_PER_GROUP(sb
), ino
);
485 if (ino
>= EXT2_INODES_PER_GROUP(sb
)) {
487 * Rare race: find_group_xx() decided that there were
488 * free inodes in this group, but by the time we tried
489 * to allocate one, they're all gone. This can also
490 * occur because the counters which find_group_orlov()
491 * uses are approximate. So just go and search the
494 if (++group
== sbi
->s_groups_count
)
498 if (ext2_set_bit_atomic(sb_bgl_lock(sbi
, group
),
499 ino
, bitmap_bh
->b_data
)) {
500 /* we lost this inode */
501 if (++ino
>= EXT2_INODES_PER_GROUP(sb
)) {
502 /* this group is exhausted, try next group */
503 if (++group
== sbi
->s_groups_count
)
507 /* try to find free inode in the same group */
508 goto repeat_in_this_group
;
514 * Scanned all blockgroups.
519 mark_buffer_dirty(bitmap_bh
);
520 if (sb
->s_flags
& MS_SYNCHRONOUS
)
521 sync_dirty_buffer(bitmap_bh
);
524 ino
+= group
* EXT2_INODES_PER_GROUP(sb
) + 1;
525 if (ino
< EXT2_FIRST_INO(sb
) || ino
> le32_to_cpu(es
->s_inodes_count
)) {
526 ext2_error (sb
, "ext2_new_inode",
527 "reserved inode or inode > inodes count - "
528 "block_group = %d,inode=%lu", group
,
529 (unsigned long) ino
);
534 percpu_counter_add(&sbi
->s_freeinodes_counter
, -1);
536 percpu_counter_inc(&sbi
->s_dirs_counter
);
538 spin_lock(sb_bgl_lock(sbi
, group
));
539 le16_add_cpu(&gdp
->bg_free_inodes_count
, -1);
541 if (sbi
->s_debts
[group
] < 255)
542 sbi
->s_debts
[group
]++;
543 le16_add_cpu(&gdp
->bg_used_dirs_count
, 1);
545 if (sbi
->s_debts
[group
])
546 sbi
->s_debts
[group
]--;
548 spin_unlock(sb_bgl_lock(sbi
, group
));
551 mark_buffer_dirty(bh2
);
552 if (test_opt(sb
, GRPID
)) {
553 inode
->i_mode
= mode
;
554 inode
->i_uid
= current_fsuid();
555 inode
->i_gid
= dir
->i_gid
;
557 inode_init_owner(inode
, dir
, mode
);
561 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME_SEC
;
562 memset(ei
->i_data
, 0, sizeof(ei
->i_data
));
564 ext2_mask_flags(mode
, EXT2_I(dir
)->i_flags
& EXT2_FL_INHERITED
);
571 ei
->i_block_alloc_info
= NULL
;
572 ei
->i_block_group
= group
;
573 ei
->i_dir_start_lookup
= 0;
574 ei
->i_state
= EXT2_STATE_NEW
;
575 ext2_set_inode_flags(inode
);
576 spin_lock(&sbi
->s_next_gen_lock
);
577 inode
->i_generation
= sbi
->s_next_generation
++;
578 spin_unlock(&sbi
->s_next_gen_lock
);
579 if (insert_inode_locked(inode
) < 0) {
584 dquot_initialize(inode
);
585 err
= dquot_alloc_inode(inode
);
589 err
= ext2_init_acl(inode
, dir
);
593 err
= ext2_init_security(inode
,dir
);
597 mark_inode_dirty(inode
);
598 ext2_debug("allocating inode %lu\n", inode
->i_ino
);
599 ext2_preread_inode(inode
);
603 dquot_free_inode(inode
);
607 inode
->i_flags
|= S_NOQUOTA
;
609 unlock_new_inode(inode
);
614 make_bad_inode(inode
);
619 unsigned long ext2_count_free_inodes (struct super_block
* sb
)
621 struct ext2_group_desc
*desc
;
622 unsigned long desc_count
= 0;
626 struct ext2_super_block
*es
;
627 unsigned long bitmap_count
= 0;
628 struct buffer_head
*bitmap_bh
= NULL
;
630 es
= EXT2_SB(sb
)->s_es
;
631 for (i
= 0; i
< EXT2_SB(sb
)->s_groups_count
; i
++) {
634 desc
= ext2_get_group_desc (sb
, i
, NULL
);
637 desc_count
+= le16_to_cpu(desc
->bg_free_inodes_count
);
639 bitmap_bh
= read_inode_bitmap(sb
, i
);
643 x
= ext2_count_free(bitmap_bh
, EXT2_INODES_PER_GROUP(sb
) / 8);
644 printk("group %d: stored = %d, counted = %u\n",
645 i
, le16_to_cpu(desc
->bg_free_inodes_count
), x
);
649 printk("ext2_count_free_inodes: stored = %lu, computed = %lu, %lu\n",
650 percpu_counter_read(&EXT2_SB(sb
)->s_freeinodes_counter
),
651 desc_count
, bitmap_count
);
654 for (i
= 0; i
< EXT2_SB(sb
)->s_groups_count
; i
++) {
655 desc
= ext2_get_group_desc (sb
, i
, NULL
);
658 desc_count
+= le16_to_cpu(desc
->bg_free_inodes_count
);
664 /* Called at mount-time, super-block is locked */
665 unsigned long ext2_count_dirs (struct super_block
* sb
)
667 unsigned long count
= 0;
670 for (i
= 0; i
< EXT2_SB(sb
)->s_groups_count
; i
++) {
671 struct ext2_group_desc
*gdp
= ext2_get_group_desc (sb
, i
, NULL
);
674 count
+= le16_to_cpu(gdp
->bg_used_dirs_count
);